Meeting global mental health needs, with technology’s help

The World Health Organization (WHO) estimates that nearly 1 billion people are living with a mental disorder, worldwide. During the global pandemic, the world saw a 25% increase in the prevalence of anxiety and depression. There was even a corresponding spike in searches on Google for mental health resources — which is a trend that continues to climb each year. To help people connect with timely, life-saving information and resources and to empower them to take action on their mental health needs, teams of Googlers are working — inside and outside of the company — to make sure everyone has access to mental health support.

Connecting people to resources on Search and YouTube

Before we can connect people to timely information and resources, we need to understand their intent when they turn to Search. Earlier this year, we shared our goal to automatically and more accurately detect personal crisis searches on Google Search, with the help of AI. This week, we’re rolling out this capability across the globe. This change enables us to better understand if someone is in crisis, then present them with reliable, actionable information. Over the coming months, we’ll work with partners to identify national suicide hotlines and make these resources accessible in dozens more languages.

Beyond the immediate needs related to mental health crises, people want information along their mental health journey no matter what it looks like — including content that can help them connect with others with similar experiences. To better support these needs, YouTube recently launched its Personal Stories feature, which surfaces content from creators who share personal experiences and stories about health topics, including anxiety, depression, post-traumatic stress disorder, addiction, bipolar disease, schizophrenia and obsessive-compulsive disorder. This feature is currently available in the U.S., with plans to expand it to more regions and to cover more health issues.

Scaling an LGBTQ+ helpline to support teens around the world

Mental health challenges are particularly prevalent in the LGBTQ+ youth community, with 45% of LGBTQ+ youth reporting that they have seriously considered attempting suicide in the past year. Since 2019, Google.org has given $2.7 million to support the work of The Trevor Project, the world’s largest suicide prevention and mental health organization for LGBTQ+ young people. With the help of a technical team of Google.org Fellows, The Trevor Project built an AI system that could identify and prioritize high-risk contacts while simultaneously reaching more LGBTQ+ young people in crisis.

Today, we’re granting $2 million to The Trevor Project to help them to scale their digital crisis services to more countries, starting with Mexico. With this funding, they will continue to build and optimize a platform to help them more quickly scale their life-affirming services globally. In addition, we’ll provide volunteer support from Google’s AI experts and $500,000 in donated Search advertising to help connect young people to these valuable resources. The Trevor Project hopes that this project will help them reach more than 40 million LGBTQ+ young people worldwide who seriously consider suicide each year.

Using AI-powered tools to provide mental health support for the veteran community

That’s not the only way The Trevor Project has tapped AI to help support their mission. Last year, with the help of Google.org Fellows, they built a Crisis Contact Simulator that has helped them train thousands of counselors. Thanks to this tool, they can increase the capacity of their highly trained crisis counselors while decreasing the human effort required for training.

Now we’re supporting ReflexAI, an organization focused on building AI-powered public safety and crisis intervention tools, to develop a similar crisis simulation technology for the veteran community. The Department of Veteran Affairs reports that more than 6,000 veterans die by suicide each year. ReflexAI will receive a team of Google.org Fellows working full-time pro bono to help the organization build a training and simulation tool for veterans so they can better support each other and encourage their peers to seek additional support when needed.

Perhaps the most potent element of all, in an effective crisis service system, is relationships. To be human. To be compassionate. We know from experience that immediate access to help, hope and healing saves lives. SAMHSA
(Substance Abuse & Mental Health Services Admin.)

When it comes to mental health, the most important path forward is connection. AI and other technologies can provide timely, life-saving resources, but the goal of all these projects is to connect people to people.

Note: Source for SAMSA quote

Read More

Beyond Words: Large Language Models Expand AI’s Horizon

Back in 2018, BERT got people talking about how machine learning models were learning to read and speak. Today, large language models, or LLMs, are growing up fast, showing dexterity in all sorts of applications.

They’re, for one, speeding drug discovery, thanks to research from the Rostlab at Technical University of Munich, as well as work by a team from Harvard, Yale and New York University and others. In separate efforts, they applied LLMs to interpret the strings of amino acids that make up proteins, advancing our understanding of these building blocks of biology.

It’s one of many inroads LLMs are making in healthcare, robotics and other fields.

A Brief History of LLMs

Transformer models — neural networks, defined in 2017, that can learn context in sequential data — got LLMs started.

Researchers behind BERT and other transformer models made 2018 “a watershed moment” for natural language processing, a report on AI said at the end of that year. “Quite a few experts have claimed that the release of BERT marks a new era in NLP,” it added.

Developed by Google, BERT (aka Bidirectional Encoder Representations from Transformers) delivered state-of-the-art scores on benchmarks for NLP. In 2019, it announced BERT powers the company’s search engine.

Google released BERT as open-source software, spawning a family of follow-ons and setting off a race to build ever larger, more powerful LLMs.

For instance, Meta created an enhanced version called RoBERTa, released as open-source code in July 2017. For training, it used “an order of magnitude more data than BERT,” the paper said, and leapt ahead on NLP leaderboards. A scrum followed.

Scaling Parameters and Markets

For convenience, score is often kept by the number of an LLM’s parameters or weights, measures of the strength of a connection between two nodes in a neural network. BERT had 110 million, RoBERTa had 123 million, then BERT-Large weighed in at 354 million, setting a new record, but not for long.

Compute required for training LLMs
As LLMs expanded into new applications, their size and computing requirements grew.

In 2020, researchers at OpenAI and Johns Hopkins University announced GPT-3, with a whopping 175 billion parameters, trained on a dataset with nearly a trillion words. It scored well on a slew of language tasks and even ciphered three-digit arithmetic.

“Language models have a wide range of beneficial applications for society,” the researchers wrote.

Experts Feel ‘Blown Away’

Within weeks, people were using GPT-3 to create poems, programs, songs, websites and more. Recently, GPT-3 even wrote an academic paper about itself.

“I just remember being kind of blown away by the things that it could do, for being just a language model,” said Percy Liang, a Stanford associate professor of computer science, speaking in a podcast.

GPT-3 helped motivate Stanford to create a center Liang now leads, exploring the implications of what it calls foundational models that can handle a wide variety of tasks well.

Toward Trillions of Parameters

Last year, NVIDIA announced the Megatron 530B LLM that can be trained for new domains and languages. It debuted with tools and services for training language models with trillions of parameters.

“Large language models have proven to be flexible and capable … able to answer deep domain questions without specialized training or supervision,” Bryan Catanzaro, vice president of applied deep learning research at NVIDIA, said at that time.

Making it even easier for users to adopt the powerful models, the NVIDIA Nemo LLM service debuted in September at GTC. It’s an NVIDIA-managed cloud service to adapt pretrained LLMs to perform specific tasks.

Transformers Transform Drug Discovery

The advances LLMs are making with proteins and chemical structures are also being applied to DNA.

Researchers aim to scale their work with NVIDIA BioNeMo, a software framework and cloud service to generate, predict and understand biomolecular data. Part of the NVIDIA Clara Discovery collection of frameworks, applications and AI models for drug discovery, it supports work in widely used protein, DNA and chemistry data formats.

NVIDIA BioNeMo features multiple pretrained AI models, including the MegaMolBART model, developed by NVIDIA and AstraZeneca.

LLM use cases in healthcare
In their paper on foundational models, Stanford researchers projected many uses for LLMs in healthcare.

LLMs Enhance Computer Vision

Transformers are also reshaping computer vision as powerful LLMs replace traditional convolutional AI models. For example, researchers at Meta AI and Dartmouth designed TimeSformer, an AI model that uses transformers to analyze video with state-of-the-art results.

Experts predict such models could spawn all sorts of new applications in computational photography, education and interactive experiences for mobile users.

In related work earlier this year, two companies released powerful AI models to generate images from text.

OpenAI announced DALL-E 2, a transformer model with 3.5 billion parameters designed to create realistic images from text descriptions. And recently, Stability AI, based in London, launched Stability Diffusion,

Writing Code, Controlling Robots

LLMs also help developers write software. Tabnine — a member of NVIDIA Inception, a program that nurtures cutting-edge startups — claims it’s automating up to 30% of the code generated by a million developers.

Taking the next step, researchers are using transformer-based models to teach robots used in manufacturing, construction, autonomous driving and personal assistants.

For example, DeepMind developed Gato, an LLM that taught a robotic arm how to stack blocks. The 1.2-billion parameter model was trained on more than 600 distinct tasks so it could be useful in a variety of modes and environments, whether playing games or animating chatbots.

Gato LLM has many applications
The Gato LLM can analyze robot actions and images as well as text.

“By scaling up and iterating on this same basic approach, we can build a useful general-purpose agent,” researchers said in a paper posted in May.

It’s another example of what the Stanford center in a July paper called a paradigm shift in AI. “Foundation models have only just begun to transform the way AI systems are built and deployed in the world,” it said.

Learn how companies around the world are implementing LLMs with NVIDIA Triton for many use cases.

The post Beyond Words: Large Language Models Expand AI’s Horizon appeared first on NVIDIA Blog.

Read More

How undesired goals can arise with correct rewards

As we build increasingly advanced artificial intelligence (AI) systems, we want to make sure they don’t pursue undesired goals. Such behaviour in an AI agent is often the result of specification gaming – exploiting a poor choice of what they are rewarded for. In our latest paper, we explore a more subtle mechanism by which AI systems may unintentionally learn to pursue undesired goals: goal misgeneralisation (GMG). GMG occurs when a system’s capabilities generalise successfully but its goal does not generalise as desired, so the system competently pursues the wrong goal. Crucially, in contrast to specification gaming, GMG can occur even when the AI system is trained with a correct specification.Read More

How undesired goals can arise with correct rewards

As we build increasingly advanced artificial intelligence (AI) systems, we want to make sure they don’t pursue undesired goals. Such behaviour in an AI agent is often the result of specification gaming – exploiting a poor choice of what they are rewarded for. In our latest paper, we explore a more subtle mechanism by which AI systems may unintentionally learn to pursue undesired goals: goal misgeneralisation (GMG). GMG occurs when a system’s capabilities generalise successfully but its goal does not generalise as desired, so the system competently pursues the wrong goal. Crucially, in contrast to specification gaming, GMG can occur even when the AI system is trained with a correct specification.Read More

AudioLM: a Language Modeling Approach to Audio Generation

AudioLM: a Language Modeling Approach to Audio Generation

Posted by Zalán Borsos, Research Software Engineer, and Neil Zeghidour, Research Scientist, Google Research

Generating realistic audio requires modeling information represented at different scales. For example, just as music builds complex musical phrases from individual notes, speech combines temporally local structures, such as phonemes or syllables, into words and sentences. Creating well-structured and coherent audio sequences at all these scales is a challenge that has been addressed by coupling audio with transcriptions that can guide the generative process, be it text transcripts for speech synthesis or MIDI representations for piano. However, this approach breaks when trying to model untranscribed aspects of audio, such as speaker characteristics necessary to help people with speech impairments recover their voice, or stylistic components of a piano performance.

In “AudioLM: a Language Modeling Approach to Audio Generation”, we propose a new framework for audio generation that learns to generate realistic speech and piano music by listening to audio only. Audio generated by AudioLM demonstrates long-term consistency (e.g., syntax in speech, melody in music) and high fidelity, outperforming previous systems and pushing the frontiers of audio generation with applications in speech synthesis or computer-assisted music. Following our AI Principles, we’ve also developed a model to identify synthetic audio generated by AudioLM.

From Text to Audio Language Models

In recent years, language models trained on very large text corpora have demonstrated their exceptional generative abilities, from open-ended dialogue to machine translation or even common-sense reasoning. They have further shown their capacity to model other signals than texts, such as natural images. The key intuition behind AudioLM is to leverage such advances in language modeling to generate audio without being trained on annotated data.

However, some challenges need to be addressed when moving from text language models to audio language models. First, one must cope with the fact that the data rate for audio is significantly higher, thus leading to much longer sequences — while a written sentence can be represented by a few dozen characters, its audio waveform typically contains hundreds of thousands of values. Second, there is a one-to-many relationship between text and audio. This means that the same sentence can be rendered by different speakers with different speaking styles, emotional content and recording conditions.

To overcome both challenges, AudioLM leverages two kinds of audio tokens. First, semantic tokens are extracted from w2v-BERT, a self-supervised audio model. These tokens capture both local dependencies (e.g., phonetics in speech, local melody in piano music) and global long-term structure (e.g., language syntax and semantic content in speech, harmony and rhythm in piano music), while heavily downsampling the audio signal to allow for modeling long sequences.

However, audio reconstructed from these tokens demonstrates poor fidelity. To overcome this limitation, in addition to semantic tokens, we rely on acoustic tokens produced by a SoundStream neural codec, which capture the details of the audio waveform (such as speaker characteristics or recording conditions) and allow for high-quality synthesis. Training a system to generate both semantic and acoustic tokens leads simultaneously to high audio quality and long-term consistency.

Training an Audio-Only Language Model

AudioLM is a pure audio model that is trained without any text or symbolic representation of music. AudioLM models an audio sequence hierarchically, from semantic tokens up to fine acoustic tokens, by chaining several Transformer models, one for each stage. Each stage is trained for the next token prediction based on past tokens, as one would train a text language model. The first stage performs this task on semantic tokens to model the high-level structure of the audio sequence.

In the second stage, we concatenate the entire semantic token sequence, along with the past coarse acoustic tokens, and feed both as conditioning to the coarse acoustic model, which then predicts the future tokens. This step models acoustic properties such as speaker characteristics in speech or timbre in music.

In the third stage, we process the coarse acoustic tokens with the fine acoustic model, which adds even more detail to the final audio. Finally, we feed acoustic tokens to the SoundStream decoder to reconstruct a waveform.

After training, one can condition AudioLM on a few seconds of audio, which enables it to generate consistent continuation. In order to showcase the general applicability of the AudioLM framework, we consider two tasks from different audio domains:

  • Speech continuation, where the model is expected to retain the speaker characteristics, prosody and recording conditions of the prompt while producing new content that is syntactically correct and semantically consistent.
  • Piano continuation, where the model is expected to generate piano music that is coherent with the prompt in terms of melody, harmony and rhythm.

In the video below, you can listen to examples where the model is asked to continue either speech or music and generate new content that was not seen during training. As you listen, note that everything you hear after the gray vertical line was generated by AudioLM and that the model has never seen any text or musical transcription, but rather just learned from raw audio. We release more samples on this webpage.

To validate our results, we asked human raters to listen to short audio clips and decide whether it is an original recording of human speech or a synthetic continuation generated by AudioLM. Based on the ratings collected, we observed a 51.2% success rate, which is not statistically significantly different from the 50% success rate achieved when assigning labels at random. This means that speech generated by AudioLM is hard to distinguish from real speech for the average listener.

Our work on AudioLM is for research purposes and we have no plans to release it more broadly at this time. In alignment with our AI Principles, we sought to understand and mitigate the possibility that people could misinterpret the short speech samples synthesized by AudioLM as real speech. For this purpose, we trained a classifier that can detect synthetic speech generated by AudioLM with very high accuracy (98.6%). This shows that despite being (almost) indistinguishable to some listeners, continuations generated by AudioLM are very easy to detect with a simple audio classifier. This is a crucial first step to help protect against the potential misuse of AudioLM, with future efforts potentially exploring technologies such as audio “watermarking”.

Conclusion

We introduce AudioLM, a language modeling approach to audio generation that provides both long-term coherence and high audio quality. Experiments on speech generation show not only that AudioLM can generate syntactically and semantically coherent speech without any text, but also that continuations produced by the model are almost indistinguishable from real speech by humans. Moreover, AudioLM goes well beyond speech and can model arbitrary audio signals such as piano music. This encourages the future extensions to other types of audio (e.g., multilingual speech, polyphonic music, and audio events) as well as integrating AudioLM into an encoder-decoder framework for conditioned tasks such as text-to-speech or speech-to-speech translation.

Acknowledgments

The work described here was authored by Zalán Borsos, Raphaël Marinier, Damien Vincent, Eugene Kharitonov, Olivier Pietquin, Matt Sharifi, Olivier Teboul, David Grangier, Marco Tagliasacchi and Neil Zeghidour. We are grateful for all discussions and feedback on this work that we received from our colleagues at Google.

Read More

Safe Real-World Reinforcement Learning for Mobile Agent Obstacle Avoidance

Collision avoidance is key for mobile robots and agents to operate safely in the real world. In this work, we present an efficient and effective collision avoidance system that combines real-world reinforcement learning (RL), search-based online trajectory planning, and automatic emergency intervention, e.g. automatic emergency braking (AEB). The goal of the RL is to learn effective search heuristics that speed up the search for collision-free trajectory and reduce the frequency of triggering automatic emergency interventions. This novel setup enables RL to learn safely and directly on mobile…Apple Machine Learning Research

Study finds the risks of sharing health care data are low

Study finds the risks of sharing health care data are low

In recent years, scientists have made great strides in their ability to develop artificial intelligence algorithms that can analyze patient data and come up with new ways to diagnose disease or predict which treatments work best for different patients.

The success of those algorithms depends on access to patient health data, which has been stripped of personal information that could be used to identify individuals from the dataset. However, the possibility that individuals could be identified through other means has raised concerns among privacy advocates.

In a new study, a team of researchers led by MIT Principal Research Scientist Leo Anthony Celi has quantified the potential risk of this kind of patient re-identification and found that it is currently extremely low relative to the risk of data breach. In fact, between 2016 and 2021, the period examined in the study, there were no reports of patient re-identification through publicly available health data.

The findings suggest that the potential risk to patient privacy is greatly outweighed by the gains for patients, who benefit from better diagnosis and treatment, says Celi. He hopes that in the near future, these datasets will become more widely available and include a more diverse group of patients.

“We agree that there is some risk to patient privacy, but there is also a risk of not sharing data,” he says. “There is harm when data is not shared, and that needs to be factored into the equation.”

Celi, who is also an instructor at the Harvard T.H. Chan School of Public Health and an attending physician with the Division of Pulmonary, Critical Care and Sleep Medicine at the Beth Israel Deaconess Medical Center, is the senior author of the new study. Kenneth Seastedt, a thoracic surgery fellow at Beth Israel Deaconess Medical Center, is the lead author of the paper, which appears today in PLOS Digital Health.

Risk-benefit analysis

Large health record databases created by hospitals and other institutions contain a wealth of information on diseases such as heart disease, cancer, macular degeneration, and Covid-19, which researchers use to try to discover new ways to diagnose and treat disease.

Celi and others at MIT’s Laboratory for Computational Physiology have created several publicly available databases, including the Medical Information Mart for Intensive Care (MIMIC), which they recently used to develop algorithms that can help doctors make better medical decisions. Many other research groups have also used the data, and others have created similar databases in countries around the world.

Typically, when patient data is entered into this kind of database, certain types of identifying information are removed, including patients’ names, addresses, and phone numbers. This is intended to prevent patients from being re-identified and having information about their medical conditions made public.

However, concerns about privacy have slowed the development of more publicly available databases with this kind of information, Celi says. In the new study, he and his colleagues set out to ask what the actual risk of patient re-identification is. First, they searched PubMed, a database of scientific papers, for any reports of patient re-identification from publicly available health data, but found none.

To expand the search, the researchers then examined media reports from September 2016 to September 2021, using Media Cloud, an open-source global news database and analysis tool. In a search of more than 10,000 U.S. media publications during that time, they did not find a single instance of patient re-identification from publicly available health data.

In contrast, they found that during the same time period, health records of nearly 100 million people were stolen through data breaches of information that was supposed to be securely stored.

“Of course, it’s good to be concerned about patient privacy and the risk of re-identification, but that risk, although it’s not zero, is minuscule compared to the issue of cyber security,” Celi says.

Better representation

More widespread sharing of de-identified health data is necessary, Celi says, to help expand the representation of minority groups in the United States, who have traditionally been underrepresented in medical studies. He is also working to encourage the development of more such databases in low- and middle-income countries.

“We cannot move forward with AI unless we address the biases that lurk in our datasets,” he says. “When we have this debate over privacy, no one hears the voice of the people who are not represented. People are deciding for them that their data need to be protected and should not be shared. But they are the ones whose health is at stake; they’re the ones who would most likely benefit from data-sharing.”

Instead of asking for patient consent to share data, which he says may exacerbate the exclusion of many people who are now underrepresented in publicly available health data, Celi recommends enhancing the existing safeguards that are in place to protect such datasets. One new strategy that he and his colleagues have begun using is to share the data in a way that it can’t be downloaded, and all queries run on it can be monitored by the administrators of the database. This allows them to flag any user inquiry that seems like it might not be for legitimate research purposes, Celi says.

“What we are advocating for is performing data analysis in a very secure environment so that we weed out any nefarious players trying to use the data for some other reasons apart from improving population health,” he says. “We’re not saying that we should disregard patient privacy. What we’re saying is that we have to also balance that with the value of data sharing.”

The research was funded by the National Institutes of Health through the National Institute of Biomedical Imaging and Bioengineering.

Read More